[MPICH] bsend behavior differences - any sample pgms ?

[Vlad Branzoi]

Sorry I mixed up the machines in my explanation.  I really don't think it's
a WinXP 32bit vs 64bit issue.  It's on the 64bit WinXP that buffered send
sends the messages as soon as possible and it's on the 32bit WinXP machine
that it only sends it out when I do a detach.  However, I did not observe
this behavior on other WinXP 32bit machines so the OS most likely has
nothing to do with it.  My mistake for confusing the issue with the OSes.
More likely it's some other factor that's influencing this and I was
wondering how MPI decides how to empty out the buffer.

As far as I've seen on this particular machine (albeit with relatively
smallish buffers - 30-50 elements of about 320K each element) it never
empties the buffer except when a detach is called.

Thanks,
Vlad

-----Original Message-----
From: Jayesh Krishna [mailto:jayesh at mcs.anl.gov]
Sent: Tuesday, January 08, 2008 15:15 PM
To: 'Vlad Branzoi'
Cc: 'mpich'
Subject: RE: [MPICH] bsend behavior differences - any sample pgms ?


 Hi,
  Do you have a sample program showing the behaviour ? Have you tried
installing the 64-bit version of MPICH2 on 64-bit version of WinXP and
running your program ?

Regards,
Jayesh

-----Original Message-----
From: owner-mpich-discuss at mcs.anl.gov
[mailto:owner-mpich-discuss at mcs.anl.gov] On Behalf Of Vlad Branzoi
Sent: Tuesday, January 08, 2008 9:59 AM
To: mpich
Subject: [MPICH] bsend behavior differences

Hi,

I have a question about the behavior of buffered send on different machines.
On one machine MPI seems to send stuff from the buffer on a consistent basis
while on another machine it only sends messages out when I 'flush' the
buffer via a buffer_detach and buffer_attach.  What could be causing this
different behavior, and is there a way to make it consistent.

Both machines are running MPICH2 1.0.6p1.  The first one is running WinXP
32-bit version and the 2nd machine is running WinXP 64-bit version, however
both are running the 32-bit version of MPICH2.

Thanks,
Vlad